Intelligent fault diagnosis system and method for rotary steering drilling tool

Abstract

The invention discloses an intelligent fault diagnosis system and method for a rotary steering drilling tool. The method comprises the following steps: S1, acquiring sensor data in a rotary steering drilling tool instrument; s2, preprocessing the sensor data; s3, constructing a CNN-LSTM-AM combined neural network fault diagnosis model in combination with a convolutional neural network, a long short-term memory network and an attention mechanism; s4, pre-training the CNN-LSTM-AM combined neural network fault diagnosis model by using the training set, and optimizing hyper-parameters of the CNN-LSTM-AM combined neural network fault diagnosis model by using the verification set and a bald eagle optimization algorithm; and S5, the neural network fault diagnosis model reads test set data, performs fault identification and diagnosis, and performs visual analysis on a result. The invention belongs to the field of petroleum and natural gas exploration, and solves the problems that intelligent diagnosis of drilling tool faults cannot be performed and the types of model faults are too few at present.

Description

Technical Field

[0001] The present invention belongs to the technical field of oil and gas exploration, and particularly relates to an intelligent fault diagnosis system and method for rotary steerable drilling tools. Background Art

[0002] The rotary steerable drilling system is an essential tool for the development of unconventional oil and gas resources, which can multiply the drilling speed, break through the horizontal section extension limit, and significantly improve the reservoir encounter rate and wellbore quality. Since the objects of oil and gas exploration and development are invisible underground rocks and fluids, the petroleum engineering has a stronger dependence on data and higher requirements for the state of drilling tools. However, after the operation of traditional rotary steerable tools, most of the drilling information management systems are of a single form, with incomplete and untimely data entry, inconsistent data formats, and no analysis function. It mainly relies on manual query and comparison, which results in the drilling data being unable to provide a working environment for the base technicians to "reduce the time for querying information and make quick and effective decisions".

[0003] The existing patent CN118503323A discloses a downhole drilling visualization intelligent analysis system. This patent mainly focuses on data analysis and visualization and cannot make intelligent diagnosis of drilling tool failures.

[0004] The paper "Research on Rotor Fault Diagnosis Method Based on Combined Moment and CNN-SE_LSTM Model - Wang Dandan" uses the CNN-LSTM-AM fault diagnosis model of convolutional neural network, SE attention mechanism and long short-term memory neural network and develops a rotor fault diagnosis system. However, this technology does not optimize the model parameters and has too few fault types, so it cannot be directly migrated and applied to the fault diagnosis of rotary steerable drilling tools. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent fault diagnosis system and method for rotary steerable drilling tools, which solves the problems that the existing technology cannot make intelligent diagnosis of drilling tool failures and the model has too few fault types and cannot be directly migrated and applied to the fault diagnosis of rotary steerable drilling tools.

[0006] The present invention adopts the following technical solutions: An intelligent fault diagnosis method for rotary steerable drilling tools, comprising the following steps:

[0007] S1. Data acquisition: Acquire the sensor data in the rotary steerable drilling tool instrument.

[0008] S2. Data preprocessing: Preprocess the sensor data, and then divide the sensor data into a training set, a validation set and a test set.

[0009] S3. Model construction: Combine a convolutional neural network, a long short-term memory network and an attention mechanism to construct a CNN-LSTM-AM combined neural network fault diagnosis model.

[0010] S4, Model Optimization: The CNN-LSTM-AM combined neural network fault diagnosis model is pre-trained using the training set, and the hyperparameters of the CNN-LSTM-AM combined neural network fault diagnosis model are optimized using the validation set and the Condor optimization algorithm.

[0011] S5, Diagnosis and Analysis: The neural network fault diagnosis model reads the test set data, performs fault identification and diagnosis, and visually analyzes the results.

[0012] Preferably, the sensor data includes: the data values of voltage, rotational speed, current, and gravitational acceleration recorded by the sensor data acquisition system when starting operations.

[0013] Preferably, the preprocessing of the rotary steerable drilling tool sensor data includes: transcoding and saving the sensor data in the rotary steerable drilling tool instrument, deleting empty data channels, filtering out useless data channels, and setting labels by combining data with fault types.

[0014] Preferably, the division ratio of the training set, validation set, and test set is 7:2:1.

[0015] Preferably, in S3, the CNN-LSTM-AM neural network fault diagnosis model includes: an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, an SE channel attention module, a first LSTM layer, a second LSTM layer, a Dropout layer, a fully connected layer, and a classification layer.

[0016] The preprocessed sensor data is input through the input layer.

[0017] Data features are extracted through the first convolutional layer, the second convolutional layer, and the third convolutional layer.

[0018] Significant features are extracted through the first pooling layer, the second pooling layer, and the third pooling layer.

[0019] The SE channel attention module is an attention mechanism that recalculates the weight of each channel during training. Based on the recalculated weight results, it enhances the weight of useful features and suppresses features with low importance; low importance is sorted by weight.

[0020] Temporal features of the signal are extracted through the first LSTM layer and the second LSTM layer.

[0021] During training, a proportion of input units are discarded through the Dropout layer to prevent overfitting.

[0022] Feature combination is performed through the fully connected layer and transformed into a vector.

[0023] Output of the final diagnosis result through the classification layer.

[0024] Preferably, the visual analysis includes: using the CNN-LSTM-AM combined neural network fault diagnosis model obtained by training optimization to diagnose based on the S2 test set data, and then evaluating the diagnosis situation of the model according to the prediction results. The evaluation of the model specifically includes: evaluating performance indicators according to the results of test verification.

[0025] Preferably, the performance indicators include precision, accuracy, F1 value, and recall rate.

[0026] The present invention also provides another technical solution: an intelligent fault diagnosis system for a rotary steerable drilling tool, including:

[0027] A data acquisition module, which acquires sensor data in a real-time rotary steerable drilling tool instrument or historical sensor data in a database, preprocesses the sensor data of the rotary steerable drilling tool, and divides the processed sensor data into a training set, a validation set, and a test set.

[0028] A model construction module, which constructs a CNN-LSTM-AM neural network fault diagnosis model, optimizes the CNN-LSTM-AM neural network fault diagnosis model using the BES algorithm, and tests the performance of the algorithm.

[0029] A fault diagnosis module, which deploys the CNN-LSTM-AM neural network fault diagnosis model to an electronic device, connects it to the data acquisition module and the data processing module, diagnoses the faults of the rotary steerable drilling tool, and outputs the diagnosis results.

[0030] The present invention also provides another technical solution: a computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, the steps in the intelligent fault diagnosis method for a rotary steerable drilling tool are implemented.

[0031] The present invention also provides another technical solution: an electronic device, including a memory, a processor, and a program stored on the memory and executable on the processor. When the processor executes the program, the steps in the intelligent fault diagnosis method for a rotary steerable drilling tool are implemented.

[0032] The beneficial effects of the present invention are: The present invention proposes a fault diagnosis method for a rotary steerable drilling tool, which realizes the feature extraction of current, voltage, rotational speed, and gravitational acceleration through operations such as transcoding, saving, and filtering of the sensor data of the rotary steerable drilling tool; the trained model can accurately judge the existing faults.

[0033] The present invention uses a CNN-LSTM-AM neural network fault diagnosis model, which integrates the technical advantages of the CNN and LSTM models and incorporates the SE attention mechanism. The CNN network extracts fault features, the SE module enhances the weights of useful features and suppresses features with low importance, and the LSTM network processes time-delay information. The combination of the three can more effectively improve the speed and accuracy of fault detection for production data in the drilling process. At the same time, the BES algorithm is introduced based on the model to optimize the hyperparameters of the complex neural network, improving the accuracy of the model.

[0034] The present invention constructs an intelligent fault diagnosis system for rotary steerable drilling tools. The system can identify power supply faults, control faults, and hardware faults of rotary steerable drilling tools, which helps to improve the working efficiency of drilling engineering, shorten the development cycle of rotary steering technology, and provide decision-making assistance for drilling technicians. Brief Description of the Drawings

[0035] Figure 1 is a schematic flow chart of the intelligent fault diagnosis method for rotary steerable drilling tools of this application;

[0036] Figure 2 is a schematic fault tree diagram provided by this application;

[0037] Figure 3 is a schematic structural diagram of the CNN-LSTM-AM combined neural network fault diagnosis model of this application;

[0038] Figure 4 is a schematic flow chart of the training and optimization of the CNN-LSTM-AM combined neural network fault diagnosis model of this application;

[0039] Figure 5 is a schematic diagram of the modules of the intelligent fault diagnosis system for rotary steerable drilling tools of this application. Detailed Embodiments

[0040] The technical solutions in the embodiments of the present application will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0041] As Figure 1 shown, the present application proposes a fault diagnosis method for rotary steerable drilling tools, including the following steps:

[0042] S1. Data acquisition: Obtain sensor data in the rotary steerable drilling tool instrument.

[0043] S2. Data preprocessing: Preprocess the data of the rotary steerable drilling tool in step S1. After preprocessing, divide the sensor data into a training set, a validation set, and a test set.

[0044] S3. Model construction: Combine the convolutional neural network (CNN), long short-term memory network (LSTM), and attention mechanism (AM) to construct a CNN-LSTM-AM combined neural network fault diagnosis model.

[0045] S4. Model optimization: Use the training set in S2 to pre-train the neural network fault diagnosis model in S3, and use the validation set and the vulture optimization algorithm (BES) to optimize the hyperparameters of the complex combined neural network.

[0046] S5. Diagnosis and analysis: The neural network fault diagnosis model reads the data in the test set of S2, performs fault identification and diagnosis, and visually analyzes the diagnosis results.

[0047] Among them, in step S1, the sensor data includes: the data values of voltage, rotation speed, current, and gravitational acceleration recorded by the sensor data acquisition system when starting the operation. An average of 60,000 data are collected per single drilling, including 184 data channels. These values all reflect whether each part of the drilling link is operating normally, whether the steering is correct, whether the attitude is correct, and whether the drilling tool is damaged, etc.

[0048] In step S2, preprocess the sensor data of the rotary steerable drilling tool in step S1. The data includes: transcoding and saving the sensor data in the rotary steerable drilling tool instrument as Excel, deleting empty data channels, filtering out useless data channels, and setting labels by combining the data with the fault types. The division ratio of the training set, validation set, and test set is 7:2:1.

[0049] As Figure 2 shown, the fault classification for the rotary steerable drilling tool data includes three major types of faults: power supply fault, control fault, and hardware fault.

[0050] The power supply fault means that the input drive flow cannot reach the working requirement, and faults in lines or components such as the turbo generator cause the tool to be unable to provide sufficient power to support the work. The specific fault labels are unstable voltage and three-phase imbalance.

[0051] The control fault means that faults such as acquisition or feedback occur in the circuit, sensor, or controller, resulting in inaccurate measurement data, abnormal transmission, etc., causing the control part to be unable to work properly for steering. The specific fault labels are communication fault and data acquisition fault.

[0052] Hardware failures may occur after the tool has been working underground for a long time. Due to factors such as wear, fatigue, high temperature, or some unexpected situations, the tool may fail and cannot work properly. The specific fault labels include high-temperature burnout and mechanical failure.

[0053] In step S3, the CNN-LSTM-AM neural network fault diagnosis model includes: an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, an SE channel attention module, a first LSTM layer, a second LSTM layer, a Dropout layer, a fully connected layer, and a classification layer.

[0054] The input layer is used to input the preprocessed sensor data.

[0055] The first convolutional layer, the second convolutional layer, and the third convolutional layer all use wide convolutional kernels to extract data features.

[0056] The size of the first convolutional layer is 16×16×32. The size of the second convolutional layer is 16×16×32.

[0057] The size of the third convolutional layer is 16×16×32.

[0058] The first pooling layer, the second pooling layer, and the third pooling layer all use max pooling to extract significant features.

[0059] The pooling windows of the first pooling layer, the second pooling layer, and the third pooling layer are all 3×2, and the strides are all 2.

[0060] The SE channel attention module is used to recalculate the weights of each channel during training, enhance the weights of useful features, and suppress features with low importance. Importance refers to the ranking by weight.

[0061] The first LSTM layer and the second LSTM layer are both used to extract the temporal features of the signal.

[0062] The number of neurons in the first LSTM layer is 64; the number of neurons in the second LSTM layer is 32.

[0063] The Dropout layer is used to prevent overfitting by the proportion of input units discarded during training.

[0064] The fully connected layer is used for feature combination and transformation into a vector.

[0065] The classification layer is used for the output of the final diagnosis result, and the activation function is Softmax.

[0066] Such as Figure 3As shown in the figure, it is a schematic diagram of the CNN-LSTM-AM model structure of the present invention. The CNN module mainly includes a convolutional layer and a pooling layer. The convolutional layer is the core module of the CNN, and its parameters are the weights of the convolutional kernels and the biases of each channel.

[0067] The mathematical expression of the convolution calculation is as follows:

[0068]

[0069] In Equation 1, is the v-th output feature map of the l-th layer; M u represents the input feature set; l represents the l-th layer of the convolutional neural network; K u is the weight of the convolutional kernel; is the u-th feature of the input feature map in the (l-1) layer; represents the convolution operation; is the bias corresponding to the convolutional kernel in the l-th layer; F(·) is the activation function, which is used to improve the non-linearity of the CNN. Usually, the Sigmoid function and the ReLU function are adopted.

[0070] The pooling layer is set after the convolutional layer. By using the method of downsampling the data, the computational amount of the parameters is reduced. Its operation formula is as follows:

[0071]

[0072] In Equation 2, represents the output after convolution in the l-th layer; is the weight; is the bias; represents the feature map of the (l-1) layer; dowm(·) represents the pooling operation; F(·) is the activation function. Max pooling and average pooling are two common pooling operations in the pooling layer, and the maximum value and the average value in the receptive field are used as the output respectively.

[0073] Specifically, the SE channel attention mechanism is an attention mechanism that can be embedded in other classification or detection models. The core lies in that the network learns the feature weights according to the loss function, so that the weights of the effective feature maps are larger, and the weights of the ineffective or less effective feature maps are smaller, so that the training model can achieve better results.

[0074] The CNN has two major characteristics of local connection and weight sharing. The SE channel attention mechanism can not only solve the problem of feature loss caused by the different proportions between channels in the CNN network, but also process the entire input feature, recalibrate the weights of the feature maps, and obtain more important feature information. During the feature extraction process, the receptive field covers the global input feature, avoiding feature loss.

[0075] Specifically, the internal operation process of the SE channel attention mechanism generally consists of three steps:

[0076] In the first step, the squeeze operation compresses features along the spatial dimension, turning each two-dimensional feature channel into a real number, and the output dimension matches the number of input feature channels. Its expression is as follows:

[0077]

[0078] In Equation 3, Z c represents the output of the squeeze operation; F sq (·) represents the squeeze operation; u c represents the c-th feature map of the input matrix; w represents the width of the input feature map; h represents the height of the input feature map; u c (i, j) represents the pixel feature at the i-th row and the j-th column.

[0079] In the second step, the excitation operation, after compressing to obtain the global features among channels, performs excitation on the global features. This process consists of a gate mechanism composed of two fully connected layers as a standard to comprehensively capture channel dependencies. Among them, the first fully connected layer is connected to the ReLU activation layer for feature dimensionality reduction and non-linear transformation, and the second fully connected layer is connected to the Sigmoid activation function to restore the original dimension and obtain the weight s. The expression is as follows:

[0080] S = F ex (z, w) = σ(g(z, W)) = σ(W 2 δ(W 1 z)) (4)

[0081] In Equation 4, S represents the output of the excitation operation, that is, the weight matrix of each feature map; F ex represents the excitation operation; W represents the weight matrix; z is the result of the previous squeeze operation; σ and δ are the Sigmoid and ReLu activation functions respectively; W 1 is a weight matrix with a dimension of C / r × C, where C represents the number of feature vectors, and r is a scaling parameter to reduce the number of channels and thus the computational amount; W 2 is also a weight matrix with a dimension of C / r × C, and the total output dimension is 1 × 1 × C; finally, after passing through the sigmoid function, the weight matrix S is obtained.

[0082] In the third step, the reweight operation multiplies the weight S of each previous channel by all elements of the corresponding channel through multiplication to enhance important features and suppress unimportant features, making the feature extraction more directional. The expression is as follows:

[0083]

[0084] In Equation 5, represents the feature map after the reweight operation; F scale represents the Scale operation; u c represents the c-th feature map of the input matrix; S c represents the output of the previous excitation operation.

[0085] Specifically, LSTM is a variant of the recurrent neural network, which adds gated units to alleviate the problems of vanishing gradients and exploding gradients during RNN training.

[0086] In LSTM, the forget gate, input gate, and output gate can delete or add information to the memory cell. The task of adding new information to the memory cell at time t requires the simultaneous participation of the forget gate and the input gate. The irrelevant information of the memory cell at the previous moment is discarded through the forget gate, and the candidate memory cells are screened by the input gate. The two together complete the update task of the memory cell; the output of the hidden layer at time t requires the joint participation of the output gate and the memory cell to complete. Its construction formula is as follows:

[0087] i (t) = σ(W i [h t-1 , x t + b i ) (6)

[0088] f (t) = σ(W f [h t-1 , x t + b f ) (7)

[0089] o (t) = σ(W 0 [h t-1 , x t + b o ) (8)

[0090] c′ (t) = tanh(W c [h (t-1) , x t + b c ) (9)

[0091] c (t) = f (t) ⊙ c (t-1) + i (t) ⊙ c′ (t) (10)

[0092] h(t) = o (t) ☉tanh(c (t) ) (11)

[0093] In Equations (6) to (11), t represents the time; i (t) represents the input gate at time t, f (t) and o (t) respectively represent the results of the forget gate and the output gate at time t; c (t-1) is the memory cell at the previous time, c' (t) and c (t) are respectively the candidate memory cell at the current time and the memory cell at the current time; h t-1 is the final output at the previous time, x t is the input at the current time, h (t) is the final output at the current time; W f and b f , W i and b i , W o and b o , W c and b c are respectively the weights and biases of the forget gate, the input gate, the output gate, and the memory cell; σ is the activation function Sigmoid; tanh is the activation function; ⊙ is the matrix product operation.

[0094] In step S4, the model is pre-trained using the training set, and the hyperparameters of the complex combined neural network are optimized using the validation set and the Bald Eagle Optimization Algorithm (BES).

[0095] The BES algorithm is an optimization algorithm based on swarm intelligence, which is mainly divided into three stages: The first stage is the selection stage of the search space, and the bald eagle will preferentially select the space with more prey to implement the search; The second stage is the search stage, and the appropriate prey is found through behaviors such as hovering in the space; The third stage is the dive capture stage, and the bald eagle selects the best capture point for a dive and finally launches a dive towards the prey.

[0096] Specifically, in the selection stage, the bald eagle selects the search space according to the position and density of the prey, and the bald eagle in this stage updates its position using the following formula:

[0097] P i,new = P best + α·r and ·(P mean - P i ) (12)

[0098] In Equation (12), P i,new is the new position of the i-th bald eagle in this stage; P bestis the current optimal position in the search space; α is the control coefficient, and its value range is [1.5, 2]; r and is a random number between (0, 1); P mean is the average position of all individuals in the vulture population; P i is the current position of the i-th vulture.

[0099] Specifically, in the search stage, after selecting the search space, the vultures will start searching for prey in the search space and looking for the best diving position to capture the prey. The mathematical model of this stage is as follows:

[0100] θ 1 (i) = α·π·r and (13)

[0101] γ 1 (i) = θ 1 (i) + R·r and (14)

[0102]

[0103] In Equations 12 to 16, θ 1 (i) is the polar angle when the i-th vulture is flying in a spiral in the air; γ 1 (i) is the polar radius when the i-th vulture is flying in a spiral in the air; α and R are deformation coefficients, and their functions are to determine the search angle and search frequency respectively, where: α ∈ (5, 10), R ∈ (0.5, 2); x 1 (i), y 1 (i) are the coordinate positions of the i-th vulture in polar coordinates during the prey search stage.

[0104] During the search stage, the position update formula of the vulture individual is as follows:

[0105] P′ i,new = P i + x 1 (i)·(P i - P mean ) + y 1 (i)·(P i - P i+1 )α·r and ·(P mean - P i ) (17)

[0106] In Equation 16, P′ i,new is the new position of the i-th vulture during the prey search stage; P i+1 is the position of the i-th vulture after the next update.

[0107] In the specific diving capture stage, after the vulture locates the prey, it will capture the prey by diving, and other individuals in the vulture population also capture the prey in this way. The mathematical model of this stage is as follows:

[0108] θ 2 (i) = α·π·r and (18)

[0109] γ 2 (i) = θ 2 (i) (19)

[0110]

[0111] In Equations 18 to 27, θ 2 (i) is the polar angle of the i-th vulture in the diving stage; γ 2 (i) is the polar radius of the i-th vulture in the diving stage; x 2 (i), y 2 (i) is the coordinate position of the i-th vulture in the polar coordinates in the diving stage.

[0112] In the diving stage, the position update formula of the vulture individual is as follows:

[0113] P″ i,new = r and P best + x 2 (i)·(P i - c 1 P mean ) + y 2 (i)·(P i - c 2 P best ) (22)

[0114] In Equation 22, P″ i,new is the new position of the i-th vulture in the diving stage; c 1 , c 2 is the vulture movement intensity, and the value range is (1, 2).

[0115] In step S5, the test set is input into the trained model for fault identification and diagnosis, and the diagnosis results are visually analyzed, including diagnosing according to the test set data using the trained and optimized fault diagnosis model, and then evaluating the diagnosis situation of the model according to the prediction results. Specifically, it includes evaluating the performance indicators according to the results of the test verification. The performance indicators include precision, accuracy, F1 value, and recall rate.

[0116] As Figure 4 shown, the training and optimization process of the CNN-LSTM-AM combined neural network fault diagnosis model of this application is as follows:

[0117] T1: Obtain data values such as voltage, rotational speed, current, and gravitational acceleration from the sensor acquisition system.

[0118] T2: Preprocess the sensor data, and after processing, it is divided into: training set, validation set, and test set.

[0119] T3: Construct the CNN-LSTM-AM neural network structure, initialize the BES algorithm, set the main parameters, and use the weight parameters as the prey.

[0120] T4: Select the search space, calculate the fitness value of the condor individuals in the search space, and determine the current optimal position.

[0121] T5: Dive to capture the prey. After the condor reaches the optimal dive position, it captures the prey by diving, updates the position of the condor at this stage, and simultaneously updates the current optimal solution.

[0122] T6: Determine whether the number of iterations of the BES algorithm has reached the maximum. If it has reached the maximum, output the threshold and the optimal value of the weight parameters; otherwise, continue the iteration.

[0123] T7: Determine the parameters of the CNN-LSTM-AM combined neural network fault diagnosis model, train and output the model.

[0124] As Figure 5 shown, the present application provides a module for an intelligent fault diagnosis system of a rotary steerable drilling tool, including:

[0125] A data acquisition module, which acquires sensor data in a real-time rotary steerable drilling tool instrument or historical sensor data in a database, preprocesses the rotary steerable drilling tool sensor data, and after processing, divides it into a training set, a validation set, and a test set.

[0126] A model construction module, which constructs a CNN-LSTM-AM neural network fault diagnosis model, uses the BES algorithm to optimize the CNN-LSTM-AM neural network fault diagnosis model, and tests the algorithm performance.

[0127] A fault diagnosis module, which deploys the CNN-LSTM-AM neural network fault diagnosis model to an electronic device, connects it to the data acquisition module and the data processing module, performs fault diagnosis on the rotary steerable drilling tool, and outputs the diagnosis result.

[0128] The implementation process of the module for the intelligent fault diagnosis system of the rotary steerable drilling tool in the present application is the steps of the intelligent fault diagnosis method for the rotary steerable drilling tool.

[0129] The intelligent fault diagnosis system for rotary steerable drilling tools constructed in this application can identify power supply faults, control faults, and hardware faults of rotary steerable drilling tools, which helps improve the working efficiency of drilling engineering, shorten the development cycle of rotary steering technology, and provide decision-making assistance for drilling technicians.

[0130] Although the content of this application has been introduced in detail through the above preferred embodiments, it should be recognized that the above description should not be regarded as a limitation of this application. After those skilled in the art have read the above content, various modifications and alternatives to this application will be obvious. Therefore, the protection scope of this application should be defined by the appended claims.

Claims

1. A method for intelligent diagnosis of rotary steerable drilling tool faults, characterized in that: The steps include: S1. Data acquisition: acquiring sensor data in the rotary steerable drilling tool instrument; S2, data preprocessing: preprocessing the sensor data, and then dividing the sensor data into a training set, a validation set, and a test set; S3, model construction: combining convolutional neural network, long short-term memory network and attention mechanism to build a CNN-LSTM-AM combined neural network fault diagnosis model; S4, model optimization: pre-training the CNN-LSTM-AM combined neural network fault diagnosis model using the training set, and optimizing the hyperparameters of the CNN-LSTM-AM combined neural network fault diagnosis model using the validation set and the vulture optimization algorithm; S5. Diagnosis and analysis: The neural network fault diagnosis model reads the test set data, performs fault identification and diagnosis, and performs visual analysis on the results.

2. The intelligent diagnosis method for rotary steerable drilling tool fault according to claim 1, characterized in that: The sensor data includes: data values ​​of voltage, rotation speed, current and gravity acceleration recorded by the sensor data acquisition system when starting the operation.

3. The intelligent diagnosis method for rotary steerable drilling tool fault according to claim 1, characterized in that: Preprocessing the rotary steerable drilling tool sensor data includes: transcoding and saving the sensor data in the rotary steerable drilling tool instrument, deleting empty data channels, filtering out useless data channels, and setting labels in combination with data and fault types.

4. The intelligent diagnosis method for rotary steerable drilling tool fault according to claim 1, characterized in that: The training set, validation set, and test set are divided in a ratio of 7:2:

1.

5. The intelligent diagnosis method for rotary steerable drilling tool fault according to claim 1, characterized in that: In S3, the CNN-LSTM-AM neural network fault diagnosis model includes: an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, an SE channel attention module, a first LSTM layer, a second LSTM layer, a Dropout layer, a fully connected layer and a classification layer; Inputting the preprocessed sensor data through the input layer; Extracting data features through the first convolution layer, the second convolution layer and the third convolution layer; Extracting salient features through the first pooling layer, the second pooling layer and the third pooling layer; The SE channel attention module is an attention mechanism, which uses the SE channel attention module to recalculate the weight of each channel during the training process, and according to the recalculated weight results, the weight of useful features is increased and the features with low importance are suppressed; the low importance is sorted by weight; Extracting the temporal features of the signal through the first LSTM layer and the second LSTM layer; The Dropout layer discards the input unit ratio during training to prevent overfitting; The features are combined through the fully connected layer and converted into a vector; The final diagnosis result is output through the classification layer.

6. The intelligent diagnosis method for rotary steerable drilling tool fault according to claim 1, characterized in that: The visual analysis includes: using the CNN-LSTM-AM combined neural network fault diagnosis model obtained through training optimization to perform diagnosis based on the test set data described in S2, and then evaluating the diagnosis status of the model based on the prediction results. Specifically, the evaluation model evaluates performance indicators based on the results of test verification.

7. The intelligent diagnosis method for rotary steerable drilling tool fault according to claim 6, characterized in that: The performance indicators include precision, accuracy, F1 value and recall rate.

8. A system using the intelligent diagnosis method for rotary steerable drilling tool faults according to any one of claims 1 to 7, characterized in that: include: A data acquisition module, which acquires sensor data from a real-time rotary steerable drilling tool instrument or historical sensor data from a database, preprocesses the rotary steerable drilling tool sensor data, and divides the processed sensor data into a training set, a validation set, and a test set; Model building module, build the CNN-LSTM-AM neural network fault diagnosis model, use the BES algorithm to optimize the CNN-LSTM-AM neural network fault diagnosis model, and test the algorithm performance; The fault diagnosis module deploys the CNN-LSTM-AM neural network fault diagnosis model into the electronic equipment, and is connected with the data acquisition module and the data processing module to perform fault diagnosis on the rotary steerable drilling tool and output the diagnosis results.

9. A computer-readable storage medium storing a program, characterized in that: When the program is executed by a processor, the steps of the intelligent diagnosis method for rotary steerable drilling tool faults as described in any one of claims 1 to 7 are implemented.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the intelligent diagnosis method for rotary steerable drilling tool faults as described in any one of claims 1-7 are implemented.

Images